Sleep-based neuronal oscillations as a physiological biomarker for Alzheimer's disease: is night time the right time?

Dementia is a devastating syndrome characterized by memory problems, confusion, and behavior changes that prevent people from performing everyday activities. It is the leading cause of dependency and disability amongst older people, resulting in a global economic cost of ~$1.3 trillion. Around 58 million people are currently afflicted by dementia worldwide, with Alzheimer's disease (AD) causing most (60–70%) of these cases. Clinical trials for novel disease-modifying therapies for AD have repeatedly been unsuccessful over the past few decades, despite showing great promise in preclinical studies. However, some optimism has been restored in the last year due to the reported success of two novel biologic therapies in phase 3 clinical trials, lecanameb (Dyck et al., 2023) and donanemab (Sims et al., 2023), which were able to significantly slow disease progression in the prodromal phase of AD. Both treatments are monoclonal antibodies that target amyloid-β (Aβ), a protein that aggregates into the extracellular plaques that begin to form in the earliest stages of the disease. These new disease-modifying interventions highlight the key importance that early diagnosis will have if we are to reach the goal of turning AD into a treatable condition.

Early diagnosis of AD will be underpinned by the discovery of biomarkers of disease that are specific enough to distinguish early AD from other memory disorders, sensitive enough to detect the very earliest stages of the disease, and cost-effective enough to allow mass screening at the population level. This will likely be best enabled by combining markers of pathology (e.g., measuring Aβ levels in blood, cerebrospinal fluid, or directly in the brain via positron emission tomography imaging) with other physiological and cognitive measures. It is particularly important to facilitate accurate early diagnosis, given that the potential side effects of both non-disease modifying and recent disease modifying treatments are significant, as are the psychological and financial impacts of AD diagnosis for patients and their families.

Measuring the brain's electrical activity provides an attractive strategy for developing functional biomarkers for AD. This can be readily done using electroencephalography (EEG), a non-invasive method that employs electrodes placed on the scalp. EEG has been used for decades as a research and diagnostic tool in clinical settings. This includes its use in support of dementia diagnosis, although these efforts are hampered by a lack of clear and reliable assessment criteria (Lee et al., 2015). Recent developments in EEG technology make ambulatory EEG, in which a wearable device is used to capture brain activity throughout the day and night as people perform their normal activities, increasingly feasible at low cost in the home (reviewed by Biondi et al., 2022). Indeed, recent studies have found that portable multichannel EEGs, designed to be quickly worn by non-experts, can provide comparable EEG data to conventional systems used in a hospital setting (e.g., Kamousi et al., 2019). This would enable sampling of a wide repertoire of disease-relevant patterns of brain activity in a manner that is minimally obtrusive and that could be deployed at a meaningful scale, thereby creating opportunities for EEG to be used as a diagnostic tool for AD if suitable assessment criteria can be defined.

Which electrical signals might be useful markers of brain function and health? Specific cognitive and behavioral states are reliably associated with different patterns of electrical activity in the brain, called neuronal oscillations. For example, gamma oscillations occur at ~30–120 Hz and are observed in multiple brain areas during periods of high cognitive load (e.g. problem solving), while non-dreaming sleep is marked by three cardinal neuronal oscillations: high-frequency (~120–250 Hz) hippocampal sharp wave ripples (SWRs), cortical slow wave oscillations (SWOs; < 1 Hz), and spindle oscillations (~8–16 Hz; see Figure 1). The neural circuitry that generates each type of oscillation is distinct, and the types of neurons that drive these oscillations show differential vulnerability to different brain diseases (see Pelkey et al., 2017 for our recent review). Thus, detecting aberrant, disease-mediated changes in neuronal oscillations can provide valuable insight into the malfunction of neuronal circuits and cellular subtypes that generate them.

Figure 1.

Neuronal oscillations in non-dreaming sleep.

In humans and rodent models, slow-wave oscillations and spindles arise throughout the cerebral cortex (blue) and can be recorded using electroencephalography, while sharp wave-ripples occur in the hippocampus (red). Example traces of each oscillation type are illustrated on the right. Created with BioRender.com.

Whilst gamma oscillations have received a lot of attention in AD research, a discord currently exists within and between studies conducted in humans and mouse models (Palop and Mucke, 2016). Both reductions and increases in gamma oscillation power have been identified, with no clear pattern emerging that correlates with brain or pathological state. This may be due to the dynamic response of gamma oscillations to movement and different cognitive tasks, and a lack of standardization when recording and analyzing these oscillations. Therefore, using EEG to record and interpret gamma oscillations in ambulating humans for AD diagnosis might prove to be technically challenging due to these confounds. Sleep-associated neuronal oscillations, on the other hand, could be an appealing alternative neurophysiological measure due to the consistent behavioral state in which they occur.

Cortical SWOs, spindles, and hippocampal SWRs are important for memory. These oscillations frequently occur together during non-dreaming sleep in humans and other mammals, with SWRs nested in the troughs of spindles that themselves coincide with SWO up-states. This tripartite coupling between SWOs, spindles, and SWRs is thought to facilitate communication between hippocampal and cortical neurons that enables the long-term consolidation of memories initially stored in the hippocampus, in a process named systems consolidation (Klinzing et al., 2019). As dementia is the main phenotype associated with AD and changes in sleep architecture are commonly observed in AD patients and rodent models (Mander, 2020), these oscillations have proven attractive targets for preclinical research. Using rodent models, we and others have found hippocampal SWRs to be disrupted as a result of different AD-associated pathologies, including amyloidopathy (Brady et al., 2023), tauopathy (Witton et al., 2016) and ApoE4 (Gillespie et al., 2016). Moreover, changes to SWOs and spindles have been documented in mouse models of AD (reviewed by Katsuki et al., 2022). Importantly, similar deficits in SWOs and spindles have been measured via EEG in people with mild cognitive impairment (a clinical precursor of AD) and in cognitively normal older adults that exhibit high levels of Aβ or tau pathology (a second major pathological hallmark of AD) (reviewed by Mander, 2020), suggesting that these oscillations could be a valuable tool in linking preclinical and clinical research.

As SWRs support memory consolidation and altered SWRs have been observed in multiple rodent models of AD, deficits in these oscillations have been repeatedly proposed as a harbinger of AD-related cognitive decline and advocated as an early functional biomarker of AD (Sanchez-Aguilera and Quintanilla, 2021). Unfortunately, EEG recordings typically only allow electrical activity to be recorded from the brain's upper cortical layers, precluding the detection of oscillations located in deeper brain regions, such as hippocampal SWRs (although it may be increasingly possible to reconstruct deeper signals using modern EEG source imaging techniques). In a recent study (Brady et al., 2023), we therefore set out to test in mice whether Aβ pathology disrupts temporal coupling between hippocampal SWRs, and cortical SWOs and spindles, as these cortical oscillations can be measured using standard EEG techniques available in clinical settings. We observed that whilst mice harboring widespread cortical and hippocampal Aβ exhibited aberrant changes in SWR and spindle power, SWO-spindle-SWR coupling was not disrupted. Specifically, our findings indicated that neuronal oscillations generated in local microcircuits are most vulnerable to amyloidopathy, with SWR changes likely associated with altered local inhibition mediated by parvalbumin-expressing interneurons – a mechanism that has been proposed previously (Sanchez-Aguilera and Quintanilla, 2021).

Whilst SWO-spindle-SWR coupling was not disrupted in our study of Aβ overexpressing mice (Brady et al., 2023), impaired coupling between SWRs and spindles has been observed in another mouse model of AD that co-expresses amyloid and tau pathologies (Benthem et al., 2020). Additionally, recent human studies suggest that coupling between cortical SWOs and spindles may have utility as an AD biomarker. For example, a study that combined sleep EEG with Aβ and tau positron emission tomography imaging in cognitively normal older adults identified that SWO-spindle uncoupling could predict levels of tau but not Aβ pathology, whilst decreased SWO power correlated with levels of Aβ but not tau pathology (Winer et al., 2019). Further, a recent study in another cohort of cognitively normal older adults showed that the phase of coupling between slow SWOs (i.e., SWOs with slow down-state to up-state transitions) and spindles predicted Aβ burden (Chylinski et al., 2022). Viewed alongside our and others' findings, these observations hint towards a model in which locally generated neuronal oscillations are first disrupted in the early amyloidogenic phase of AD, with detriments in oscillatory coupling between anatomically distributed brain areas emerging later in the disease progression, potentially through the worsening of Aβ pathology and/or the development of concurrent pathologies like tauopathy. Local changes in neuronal oscillations may be caused, at least in part, by disrupted excitatory-inhibitory balance linked to the loss and/or aberrant activity of fast-spiking parvalbumin-expressing interneurons, which have been shown to exhibit amyloid pathology-related changes in excitability associated with metabolic stress and altered ion channel expression (Palop and Mucke, 2016). An appealing feature of this model is that it suggests that sleep-associated neuronal oscillations may be sufficiently sensitive to dissociate gross pathophysiological staging of AD based on whether changes to oscillation properties and/or coupling are indicative of local or long-range neuronal network dysfunction.

In summary, we suggest that cortical spindles and the SWO, measured using EEG during sleep, present attractive candidates as early-detection biomarkers of AD. In humans, SWOs and spindles can occur either globally across the entire cortex or, most often, locally in specific regions (Nir et al., 2011), so the use of multisite EEG would allow local and long-range oscillations to be disambiguated. As these oscillations occur throughout the cortex, precise recording locations matter less than the distance between electrodes to allow parsing of locally vs. globally occurring activity. While SWRs are generated in the hippocampus and thus not easily detected with scalp electrodes, combined scalp and intrahippocampal EEG recordings in humans have shown that the association between spindles and ripples occurs in a manner analogous to that observed in rodents (Ngo et al., 2020). Similarities between humans and rodent models and the potential to dissociate disease staging through assessment of local and long-range network dysfunction suggest that these oscillations could be a valuable tool that should be explored in existing datasets and prospective EEG studies (e.g., clinical trial NCT04002583) as a possible route towards bridging the translational gap that still exists between preclinical and clinical AD research.